Water leaks can cause significant damage and result in high costs for building owners, tenants, and property managers. That’s why leak detection systems have become a crucial aspect of building management. In recent years, advancements in technology have made it possible to detect leaks automatically and remotely, thanks to machine learning algorithms. But how do these algorithms work, and what are the benefits of using them?
Using machine learning for leak detection
Leak detection systems typically use machine learning algorithms to profile tenant behaviour using historical data. By analysing the volume and time of water usage during a typical weekday or weekend, the algorithm can recognize events and predict future consumption. Using the data acquired, alarm thresholds are established based on past maximum consumption events. By splitting these events by the day of the week and further dividing them by time, the algorithm can accurately detect abnormal water usage patterns and trigger an alert if necessary.
![A visual representation of a leak detection algorithm placing an envelope over a typical consumption pattern](https://cdn.prod.website-files.com/657749fbe6e6884c6503583a/657ad958b59fae33ff67163b_1684228664531-1024x318.png)
The ability to learn on-the-fly is another crucial aspect of machine learning algorithms. In the event of a false alarm, the user has the option to manually override the software and maintain the water flow. Furthermore, the algorithm will learn this activity and adapt in the future.
Is machine learning required?
While it is possible to pre-program a device based on guessed thresholds, it’s unlikely to be accurate due to changes in water consumption patterns over time. To ensure ongoing efficient and effective operation of the system throughout the building’s lifetime, remote access and machine learning are necessary components. Without online access, users lose the ability to receive remote notifications, control the valve remotely, or share access with tenants or subcontractors. Additionally, machine learning will perform ongoing performance evaluations of the building, which may encompass multiple leak prevention systems. This process guarantees that the entire system remains online and functions in seamless harmony.
![An example of minor and major alert thresholds in a leak detection algorithm](https://cdn.prod.website-files.com/657749fbe6e6884c6503583a/657ad95831367e29c64b259e_1684237300242.gif)
What happens after the system detects a leak?
After a leak is detected, a first-stage alert should provide information on the volume and duration of water usage, giving an idea of how severe the leak is. In this alert, the end user(s) can act immediately to turn the water off or over-ride the event so the water does not switch off, providing valuable feedback to the machine learning system.
![How minor and major alerts are used by the algorithm over a continuous consumption event](https://cdn.prod.website-files.com/657749fbe6e6884c6503583a/657ad95831367e29c64b258d_1684237268320-1024x570.png)
A second stage alert should then be sent if there is no response from the user, automatically shutting the water off. To find the exact location of the leak, after switching the water back on inside the app, real-time flow rate monitoring can help determine if the leak is still active and how bad it is. Usually we just want to know whether the leak is a drip, in which case we can use water temporarily until investigation, or a more severe burst. To locate the problem area, the user can switch off specific branches, tanks, and individual supplies.
![Animation showing real time monitoring in practice using FlowReporter](https://cdn.prod.website-files.com/657749fbe6e6884c6503583a/657ad959f1fb206c3225c394_1684165741333.gif)
Unfortunately flow-based monitoring is not enough to accurately pinpoint the exact location of a damaging leak without realtime monitoring and human intervention. Yes, Artificial Intelligence may be able to disaggregate events using tagging to understand how much water is used for showers/baths/toilets/taps/sprinklers etc. However, the inherent issue stems from the fact that a destructive leak will inevitably arise in an unforeseen section of the pipework, in a location that nobody has previously identified.
![FlowReporter screens showing consumption for many properties and the result of learning mode](https://cdn.prod.website-files.com/657749fbe6e6884c6503583a/657ad958477369152b327524_1684164402497-1024x576.jpeg)
Moving forward
Developing a solution to pinpoint leaks in the future has potential. For example, by including pressure data, we could solve the degrees of freedom needed to find the distance to a leak. This will of course introduce further practical complexities, such as the work needed in recording the exact layout of the mechanical infrastructure, which may become infeasible. In the meantime, we believe water contact sensors (such as leak cable or our wireless splash sensors) placed in high-risk areas solve the location problem in a much more straightforward way.
![Quensus Splash Sensor located under a wall-mounted sink](https://cdn.prod.website-files.com/657749fbe6e6884c6503583a/657ad95835cfe9f85bd10531_splash-sensor-2-287x300.png)
In conclusion, machine learning algorithms have become a crucial aspect of leak detection systems. They provide accurate and reliable monitoring of water consumption, which can help detect leaks early and prevent costly damages. Remote access and machine learning intelligence are essential for ensuring the system’s continued efficiency and effectiveness. As technology continues to evolve, we can expect more advanced algorithms to be developed, providing even better leak detection capabilities for building management.
Let us help safeguard your property against water damage, contact today.